bgpd/bgp_mpath.c

   1 /*
   2  * BGP Multipath
   3  * Copyright (C) 2010 Google Inc.
   4  *
   5  * This file is part of Quagga
   6  *
   7  * Quagga is free software; you can redistribute it and/or modify it
   8  * under the terms of the GNU General Public License as published by the
   9  * Free Software Foundation; either version 2, or (at your option) any
  10  * later version.
  11  *
  12  * Quagga is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License along
  18  * with this program; see the file COPYING; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #include <zebra.h>
  23
  24 #include "command.h"
  25 #include "prefix.h"
  26 #include "linklist.h"
  27 #include "sockunion.h"
  28 #include "memory.h"
  29 #include "queue.h"
  30 #include "filter.h"
  31
  32 #include "bgpd/bgpd.h"
  33 #include "bgpd/bgp_table.h"
  34 #include "bgpd/bgp_route.h"
  35 #include "bgpd/bgp_attr.h"
  36 #include "bgpd/bgp_debug.h"
  37 #include "bgpd/bgp_aspath.h"
  38 #include "bgpd/bgp_community.h"
  39 #include "bgpd/bgp_ecommunity.h"
  40 #include "bgpd/bgp_lcommunity.h"
  41 #include "bgpd/bgp_mpath.h"
  42
  43 /*
  44  * bgp_maximum_paths_set
  45  *
  46  * Record maximum-paths configuration for BGP instance
  47  */
  48 int bgp_maximum_paths_set(struct bgp *bgp, afi_t afi, safi_t safi, int peertype,
  49                           uint16_t maxpaths, uint16_t options)
  50 {
  51         if (!bgp || (afi >= AFI_MAX) || (safi >= SAFI_MAX))
  52                 return -1;
  53
  54         switch (peertype) {
  55         case BGP_PEER_IBGP:
  56                 bgp->maxpaths[afi][safi].maxpaths_ibgp = maxpaths;
  57                 bgp->maxpaths[afi][safi].ibgp_flags |= options;
  58                 break;
  59         case BGP_PEER_EBGP:
  60                 bgp->maxpaths[afi][safi].maxpaths_ebgp = maxpaths;
  61                 break;
  62         default:
  63                 return -1;
  64         }
  65
  66         return 0;
  67 }
  68
  69 /*
  70  * bgp_maximum_paths_unset
  71  *
  72  * Remove maximum-paths configuration from BGP instance
  73  */
  74 int bgp_maximum_paths_unset(struct bgp *bgp, afi_t afi, safi_t safi,
  75                             int peertype)
  76 {
  77         if (!bgp || (afi >= AFI_MAX) || (safi >= SAFI_MAX))
  78                 return -1;
  79
  80         switch (peertype) {
  81         case BGP_PEER_IBGP:
  82                 bgp->maxpaths[afi][safi].maxpaths_ibgp = multipath_num;
  83                 bgp->maxpaths[afi][safi].ibgp_flags = 0;
  84                 break;
  85         case BGP_PEER_EBGP:
  86                 bgp->maxpaths[afi][safi].maxpaths_ebgp = multipath_num;
  87                 break;
  88         default:
  89                 return -1;
  90         }
  91
  92         return 0;
  93 }
  94
  95 /*
  96  * bgp_interface_same
  97  *
  98  * Return true if ifindex for ifp1 and ifp2 are the same, else return false.
  99  */
 100 static int bgp_interface_same(struct interface *ifp1, struct interface *ifp2)
 101 {
 102         if (!ifp1 && !ifp2)
 103                 return 1;
 104
 105         if (!ifp1 && ifp2)
 106                 return 0;
 107
 108         if (ifp1 && !ifp2)
 109                 return 0;
 110
 111         return (ifp1->ifindex == ifp2->ifindex);
 112 }
 113
 114
 115 /*
 116  * bgp_path_info_nexthop_cmp
 117  *
 118  * Compare the nexthops of two paths. Return value is less than, equal to,
 119  * or greater than zero if bpi1 is respectively less than, equal to,
 120  * or greater than bpi2.
 121  */
 122 int bgp_path_info_nexthop_cmp(struct bgp_path_info *bpi1,
 123                               struct bgp_path_info *bpi2)
 124 {
 125         int compare;
 126         struct in6_addr addr1, addr2;
 127
 128         compare = IPV4_ADDR_CMP(&bpi1->attr->nexthop, &bpi2->attr->nexthop);
 129         if (!compare) {
 130                 if (bpi1->attr->mp_nexthop_len == bpi2->attr->mp_nexthop_len) {
 131                         switch (bpi1->attr->mp_nexthop_len) {
 132                         case BGP_ATTR_NHLEN_IPV4:
 133                         case BGP_ATTR_NHLEN_VPNV4:
 134                                 compare = IPV4_ADDR_CMP(
 135                                         &bpi1->attr->mp_nexthop_global_in,
 136                                         &bpi2->attr->mp_nexthop_global_in);
 137                                 break;
 138                         case BGP_ATTR_NHLEN_IPV6_GLOBAL:
 139                         case BGP_ATTR_NHLEN_VPNV6_GLOBAL:
 140                                 compare = IPV6_ADDR_CMP(
 141                                         &bpi1->attr->mp_nexthop_global,
 142                                         &bpi2->attr->mp_nexthop_global);
 143                                 break;
 144                         case BGP_ATTR_NHLEN_IPV6_GLOBAL_AND_LL:
 145                                 addr1 = (bpi1->attr->mp_nexthop_prefer_global)
 146                                                 ? bpi1->attr->mp_nexthop_global
 147                                                 : bpi1->attr->mp_nexthop_local;
 148                                 addr2 = (bpi2->attr->mp_nexthop_prefer_global)
 149                                                 ? bpi2->attr->mp_nexthop_global
 150                                                 : bpi2->attr->mp_nexthop_local;
 151
 152                                 if (!bpi1->attr->mp_nexthop_prefer_global
 153                                     && !bpi2->attr->mp_nexthop_prefer_global)
 154                                         compare = !bgp_interface_same(
 155                                                 bpi1->peer->ifp,
 156                                                 bpi2->peer->ifp);
 157
 158                                 if (!compare)
 159                                         compare = IPV6_ADDR_CMP(&addr1, &addr2);
 160                                 break;
 161                         }
 162                 }
 163
 164                 /* This can happen if one IPv6 peer sends you global and
 165                  * link-local
 166                  * nexthops but another IPv6 peer only sends you global
 167                  */
 168                 else if (bpi1->attr->mp_nexthop_len
 169                                  == BGP_ATTR_NHLEN_IPV6_GLOBAL
 170                          || bpi1->attr->mp_nexthop_len
 171                                     == BGP_ATTR_NHLEN_IPV6_GLOBAL_AND_LL) {
 172                         compare = IPV6_ADDR_CMP(&bpi1->attr->mp_nexthop_global,
 173                                                 &bpi2->attr->mp_nexthop_global);
 174                         if (!compare) {
 175                                 if (bpi1->attr->mp_nexthop_len
 176                                     < bpi2->attr->mp_nexthop_len)
 177                                         compare = -1;
 178                                 else
 179                                         compare = 1;
 180                         }
 181                 }
 182         }
 183
 184         return compare;
 185 }
 186
 187 /*
 188  * bgp_path_info_mpath_cmp
 189  *
 190  * This function determines our multipath list ordering. By ordering
 191  * the list we can deterministically select which paths are included
 192  * in the multipath set. The ordering also helps in detecting changes
 193  * in the multipath selection so we can detect whether to send an
 194  * update to zebra.
 195  *
 196  * The order of paths is determined first by received nexthop, and then
 197  * by peer address if the nexthops are the same.
 198  */
 199 static int bgp_path_info_mpath_cmp(void *val1, void *val2)
 200 {
 201         struct bgp_path_info *bpi1, *bpi2;
 202         int compare;
 203
 204         bpi1 = val1;
 205         bpi2 = val2;
 206
 207         compare = bgp_path_info_nexthop_cmp(bpi1, bpi2);
 208
 209         if (!compare) {
 210                 if (!bpi1->peer->su_remote && !bpi2->peer->su_remote)
 211                         compare = 0;
 212                 else if (!bpi1->peer->su_remote)
 213                         compare = 1;
 214                 else if (!bpi2->peer->su_remote)
 215                         compare = -1;
 216                 else
 217                         compare = sockunion_cmp(bpi1->peer->su_remote,
 218                                                 bpi2->peer->su_remote);
 219         }
 220
 221         return compare;
 222 }
 223
 224 /*
 225  * bgp_mp_list_init
 226  *
 227  * Initialize the mp_list, which holds the list of multipaths
 228  * selected by bgp_best_selection
 229  */
 230 void bgp_mp_list_init(struct list *mp_list)
 231 {
 232         assert(mp_list);
 233         memset(mp_list, 0, sizeof(struct list));
 234         mp_list->cmp = bgp_path_info_mpath_cmp;
 235 }
 236
 237 /*
 238  * bgp_mp_list_clear
 239  *
 240  * Clears all entries out of the mp_list
 241  */
 242 void bgp_mp_list_clear(struct list *mp_list)
 243 {
 244         assert(mp_list);
 245         list_delete_all_node(mp_list);
 246 }
 247
 248 /*
 249  * bgp_mp_list_add
 250  *
 251  * Adds a multipath entry to the mp_list
 252  */
 253 void bgp_mp_list_add(struct list *mp_list, struct bgp_path_info *mpinfo)
 254 {
 255         assert(mp_list && mpinfo);
 256         listnode_add_sort(mp_list, mpinfo);
 257 }
 258
 259 /*
 260  * bgp_path_info_mpath_new
 261  *
 262  * Allocate and zero memory for a new bgp_path_info_mpath element
 263  */
 264 static struct bgp_path_info_mpath *bgp_path_info_mpath_new(void)
 265 {
 266         struct bgp_path_info_mpath *new_mpath;
 267         new_mpath = XCALLOC(MTYPE_BGP_MPATH_INFO,
 268                             sizeof(struct bgp_path_info_mpath));
 269         return new_mpath;
 270 }
 271
 272 /*
 273  * bgp_path_info_mpath_free
 274  *
 275  * Release resources for a bgp_path_info_mpath element and zero out pointer
 276  */
 277 void bgp_path_info_mpath_free(struct bgp_path_info_mpath **mpath)
 278 {
 279         if (mpath && *mpath) {
 280                 if ((*mpath)->mp_attr)
 281                         bgp_attr_unintern(&(*mpath)->mp_attr);
 282                 XFREE(MTYPE_BGP_MPATH_INFO, *mpath);
 283                 *mpath = NULL;
 284         }
 285 }
 286
 287 /*
 288  * bgp_path_info_mpath_get
 289  *
 290  * Fetch the mpath element for the given bgp_path_info. Used for
 291  * doing lazy allocation.
 292  */
 293 static struct bgp_path_info_mpath *
 294 bgp_path_info_mpath_get(struct bgp_path_info *path)
 295 {
 296         struct bgp_path_info_mpath *mpath;
 297         if (!path->mpath) {
 298                 mpath = bgp_path_info_mpath_new();
 299                 if (!mpath)
 300                         return NULL;
 301                 path->mpath = mpath;
 302                 mpath->mp_info = path;
 303         }
 304         return path->mpath;
 305 }
 306
 307 /*
 308  * bgp_path_info_mpath_enqueue
 309  *
 310  * Enqueue a path onto the multipath list given the previous multipath
 311  * list entry
 312  */
 313 static void bgp_path_info_mpath_enqueue(struct bgp_path_info *prev_info,
 314                                         struct bgp_path_info *path)
 315 {
 316         struct bgp_path_info_mpath *prev, *mpath;
 317
 318         prev = bgp_path_info_mpath_get(prev_info);
 319         mpath = bgp_path_info_mpath_get(path);
 320         if (!prev || !mpath)
 321                 return;
 322
 323         mpath->mp_next = prev->mp_next;
 324         mpath->mp_prev = prev;
 325         if (prev->mp_next)
 326                 prev->mp_next->mp_prev = mpath;
 327         prev->mp_next = mpath;
 328
 329         SET_FLAG(path->flags, BGP_PATH_MULTIPATH);
 330 }
 331
 332 /*
 333  * bgp_path_info_mpath_dequeue
 334  *
 335  * Remove a path from the multipath list
 336  */
 337 void bgp_path_info_mpath_dequeue(struct bgp_path_info *path)
 338 {
 339         struct bgp_path_info_mpath *mpath = path->mpath;
 340         if (!mpath)
 341                 return;
 342         if (mpath->mp_prev)
 343                 mpath->mp_prev->mp_next = mpath->mp_next;
 344         if (mpath->mp_next)
 345                 mpath->mp_next->mp_prev = mpath->mp_prev;
 346         mpath->mp_next = mpath->mp_prev = NULL;
 347         UNSET_FLAG(path->flags, BGP_PATH_MULTIPATH);
 348 }
 349
 350 /*
 351  * bgp_path_info_mpath_next
 352  *
 353  * Given a bgp_path_info, return the next multipath entry
 354  */
 355 struct bgp_path_info *bgp_path_info_mpath_next(struct bgp_path_info *path)
 356 {
 357         if (!path->mpath || !path->mpath->mp_next)
 358                 return NULL;
 359         return path->mpath->mp_next->mp_info;
 360 }
 361
 362 /*
 363  * bgp_path_info_mpath_first
 364  *
 365  * Given bestpath bgp_path_info, return the first multipath entry.
 366  */
 367 struct bgp_path_info *bgp_path_info_mpath_first(struct bgp_path_info *path)
 368 {
 369         return bgp_path_info_mpath_next(path);
 370 }
 371
 372 /*
 373  * bgp_path_info_mpath_count
 374  *
 375  * Given the bestpath bgp_path_info, return the number of multipath entries
 376  */
 377 uint32_t bgp_path_info_mpath_count(struct bgp_path_info *path)
 378 {
 379         if (!path->mpath)
 380                 return 0;
 381         return path->mpath->mp_count;
 382 }
 383
 384 /*
 385  * bgp_path_info_mpath_count_set
 386  *
 387  * Sets the count of multipaths into bestpath's mpath element
 388  */
 389 static void bgp_path_info_mpath_count_set(struct bgp_path_info *path,
 390                                           uint32_t count)
 391 {
 392         struct bgp_path_info_mpath *mpath;
 393         if (!count && !path->mpath)
 394                 return;
 395         mpath = bgp_path_info_mpath_get(path);
 396         if (!mpath)
 397                 return;
 398         mpath->mp_count = count;
 399 }
 400
 401 /*
 402  * bgp_path_info_mpath_attr
 403  *
 404  * Given bestpath bgp_path_info, return aggregated attribute set used
 405  * for advertising the multipath route
 406  */
 407 struct attr *bgp_path_info_mpath_attr(struct bgp_path_info *path)
 408 {
 409         if (!path->mpath)
 410                 return NULL;
 411         return path->mpath->mp_attr;
 412 }
 413
 414 /*
 415  * bgp_path_info_mpath_attr_set
 416  *
 417  * Sets the aggregated attribute into bestpath's mpath element
 418  */
 419 static void bgp_path_info_mpath_attr_set(struct bgp_path_info *path,
 420                                          struct attr *attr)
 421 {
 422         struct bgp_path_info_mpath *mpath;
 423         if (!attr && !path->mpath)
 424                 return;
 425         mpath = bgp_path_info_mpath_get(path);
 426         if (!mpath)
 427                 return;
 428         mpath->mp_attr = attr;
 429 }
 430
 431 /*
 432  * bgp_path_info_mpath_update
 433  *
 434  * Compare and sync up the multipath list with the mp_list generated by
 435  * bgp_best_selection
 436  */
 437 void bgp_path_info_mpath_update(struct bgp_node *rn,
 438                                 struct bgp_path_info *new_best,
 439                                 struct bgp_path_info *old_best,
 440                                 struct list *mp_list,
 441                                 struct bgp_maxpaths_cfg *mpath_cfg)
 442 {
 443         uint16_t maxpaths, mpath_count, old_mpath_count;
 444         struct listnode *mp_node, *mp_next_node;
 445         struct bgp_path_info *cur_mpath, *new_mpath, *next_mpath, *prev_mpath;
 446         int mpath_changed, debug;
 447         char pfx_buf[PREFIX2STR_BUFFER], nh_buf[2][INET6_ADDRSTRLEN];
 448         char path_buf[PATH_ADDPATH_STR_BUFFER];
 449
 450         mpath_changed = 0;
 451         maxpaths = multipath_num;
 452         mpath_count = 0;
 453         cur_mpath = NULL;
 454         old_mpath_count = 0;
 455         prev_mpath = new_best;
 456         mp_node = listhead(mp_list);
 457         debug = bgp_debug_bestpath(&rn->p);
 458
 459         if (debug)
 460                 prefix2str(&rn->p, pfx_buf, sizeof(pfx_buf));
 461
 462         if (new_best) {
 463                 mpath_count++;
 464                 if (new_best != old_best)
 465                         bgp_path_info_mpath_dequeue(new_best);
 466                 maxpaths = (new_best->peer->sort == BGP_PEER_IBGP)
 467                                    ? mpath_cfg->maxpaths_ibgp
 468                                    : mpath_cfg->maxpaths_ebgp;
 469         }
 470
 471         if (old_best) {
 472                 cur_mpath = bgp_path_info_mpath_first(old_best);
 473                 old_mpath_count = bgp_path_info_mpath_count(old_best);
 474                 bgp_path_info_mpath_count_set(old_best, 0);
 475                 bgp_path_info_mpath_dequeue(old_best);
 476         }
 477
 478         if (debug)
 479                 zlog_debug(
 480                         "%s: starting mpath update, newbest %s num candidates %d old-mpath-count %d",
 481                         pfx_buf, new_best ? new_best->peer->host : "NONE",
 482                         mp_list ? listcount(mp_list) : 0, old_mpath_count);
 483
 484         /*
 485          * We perform an ordered walk through both lists in parallel.
 486          * The reason for the ordered walk is that if there are paths
 487          * that were previously multipaths and are still multipaths, the walk
 488          * should encounter them in both lists at the same time. Otherwise
 489          * there will be paths that are in one list or another, and we
 490          * will deal with these separately.
 491          *
 492          * Note that new_best might be somewhere in the mp_list, so we need
 493          * to skip over it
 494          */
 495         while (mp_node || cur_mpath) {
 496                 struct bgp_path_info *tmp_info;
 497
 498                 /*
 499                  * We can bail out of this loop if all existing paths on the
 500                  * multipath list have been visited (for cleanup purposes) and
 501                  * the maxpath requirement is fulfulled
 502                  */
 503                 if (!cur_mpath && (mpath_count >= maxpaths))
 504                         break;
 505
 506                 mp_next_node = mp_node ? listnextnode(mp_node) : NULL;
 507                 next_mpath =
 508                         cur_mpath ? bgp_path_info_mpath_next(cur_mpath) : NULL;
 509                 tmp_info = mp_node ? listgetdata(mp_node) : NULL;
 510
 511                 if (debug)
 512                         zlog_debug(
 513                                 "%s: comparing candidate %s with existing mpath %s",
 514                                 pfx_buf,
 515                                 tmp_info ? tmp_info->peer->host : "NONE",
 516                                 cur_mpath ? cur_mpath->peer->host : "NONE");
 517
 518                 /*
 519                  * If equal, the path was a multipath and is still a multipath.
 520                  * Insert onto new multipath list if maxpaths allows.
 521                  */
 522                 if (mp_node && (listgetdata(mp_node) == cur_mpath)) {
 523                         list_delete_node(mp_list, mp_node);
 524                         bgp_path_info_mpath_dequeue(cur_mpath);
 525                         if ((mpath_count < maxpaths)
 526                             && bgp_path_info_nexthop_cmp(prev_mpath,
 527                                                          cur_mpath)) {
 528                                 bgp_path_info_mpath_enqueue(prev_mpath,
 529                                                             cur_mpath);
 530                                 prev_mpath = cur_mpath;
 531                                 mpath_count++;
 532                                 if (debug) {
 533                                         bgp_path_info_path_with_addpath_rx_str(
 534                                                 cur_mpath, path_buf);
 535                                         zlog_debug(
 536                                                 "%s: %s is still multipath, cur count %d",
 537                                                 pfx_buf, path_buf, mpath_count);
 538                                 }
 539                         } else {
 540                                 mpath_changed = 1;
 541                                 if (debug) {
 542                                         bgp_path_info_path_with_addpath_rx_str(
 543                                                 cur_mpath, path_buf);
 544                                         zlog_debug(
 545                                                 "%s: remove mpath %s nexthop %s, cur count %d",
 546                                                 pfx_buf, path_buf,
 547                                                 inet_ntop(AF_INET,
 548                                                           &cur_mpath->attr
 549                                                                    ->nexthop,
 550                                                           nh_buf[0],
 551                                                           sizeof(nh_buf[0])),
 552                                                 mpath_count);
 553                                 }
 554                         }
 555                         mp_node = mp_next_node;
 556                         cur_mpath = next_mpath;
 557                         continue;
 558                 }
 559
 560                 if (cur_mpath
 561                     && (!mp_node
 562                         || (bgp_path_info_mpath_cmp(cur_mpath,
 563                                                     listgetdata(mp_node))
 564                             < 0))) {
 565                         /*
 566                          * If here, we have an old multipath and either the
 567                          * mp_list
 568                          * is finished or the next mp_node points to a later
 569                          * multipath, so we need to purge this path from the
 570                          * multipath list
 571                          */
 572                         bgp_path_info_mpath_dequeue(cur_mpath);
 573                         mpath_changed = 1;
 574                         if (debug) {
 575                                 bgp_path_info_path_with_addpath_rx_str(
 576                                         cur_mpath, path_buf);
 577                                 zlog_debug(
 578                                         "%s: remove mpath %s nexthop %s, cur count %d",
 579                                         pfx_buf, path_buf,
 580                                         inet_ntop(AF_INET,
 581                                                   &cur_mpath->attr->nexthop,
 582                                                   nh_buf[0], sizeof(nh_buf[0])),
 583                                         mpath_count);
 584                         }
 585                         cur_mpath = next_mpath;
 586                 } else {
 587                         /*
 588                          * If here, we have a path on the mp_list that was not
 589                          * previously
 590                          * a multipath (due to non-equivalance or maxpaths
 591                          * exceeded),
 592                          * or the matching multipath is sorted later in the
 593                          * multipath
 594                          * list. Before we enqueue the path on the new multipath
 595                          * list,
 596                          * make sure its not on the old_best multipath list or
 597                          * referenced
 598                          * via next_mpath:
 599                          * - If next_mpath points to this new path, update
 600                          * next_mpath to
 601                          *   point to the multipath after this one
 602                          * - Dequeue the path from the multipath list just to
 603                          * make sure
 604                          */
 605                         new_mpath = listgetdata(mp_node);
 606                         list_delete_node(mp_list, mp_node);
 607                         assert(new_mpath);
 608                         assert(prev_mpath);
 609                         if ((mpath_count < maxpaths) && (new_mpath != new_best)
 610                             && bgp_path_info_nexthop_cmp(prev_mpath,
 611                                                          new_mpath)) {
 612                                 if (new_mpath == next_mpath)
 613                                         bgp_path_info_mpath_next(new_mpath);
 614                                 bgp_path_info_mpath_dequeue(new_mpath);
 615
 616                                 bgp_path_info_mpath_enqueue(prev_mpath,
 617                                                             new_mpath);
 618                                 prev_mpath = new_mpath;
 619                                 mpath_changed = 1;
 620                                 mpath_count++;
 621                                 if (debug) {
 622                                         bgp_path_info_path_with_addpath_rx_str(
 623                                                 new_mpath, path_buf);
 624                                         zlog_debug(
 625                                                 "%s: add mpath %s nexthop %s, cur count %d",
 626                                                 pfx_buf, path_buf,
 627                                                 inet_ntop(AF_INET,
 628                                                           &new_mpath->attr
 629                                                                    ->nexthop,
 630                                                           nh_buf[0],
 631                                                           sizeof(nh_buf[0])),
 632                                                 mpath_count);
 633                                 }
 634                         }
 635                         mp_node = mp_next_node;
 636                 }
 637         }
 638
 639         if (new_best) {
 640                 if (debug)
 641                         zlog_debug(
 642                                 "%s: New mpath count (incl newbest) %d mpath-change %s",
 643                                 pfx_buf, mpath_count,
 644                                 mpath_changed ? "YES" : "NO");
 645
 646                 bgp_path_info_mpath_count_set(new_best, mpath_count - 1);
 647                 if (mpath_changed
 648                     || (bgp_path_info_mpath_count(new_best) != old_mpath_count))
 649                         SET_FLAG(new_best->flags, BGP_PATH_MULTIPATH_CHG);
 650         }
 651 }
 652
 653 /*
 654  * bgp_mp_dmed_deselect
 655  *
 656  * Clean up multipath information for BGP_PATH_DMED_SELECTED path that
 657  * is not selected as best path
 658  */
 659 void bgp_mp_dmed_deselect(struct bgp_path_info *dmed_best)
 660 {
 661         struct bgp_path_info *mpinfo, *mpnext;
 662
 663         if (!dmed_best)
 664                 return;
 665
 666         for (mpinfo = bgp_path_info_mpath_first(dmed_best); mpinfo;
 667              mpinfo = mpnext) {
 668                 mpnext = bgp_path_info_mpath_next(mpinfo);
 669                 bgp_path_info_mpath_dequeue(mpinfo);
 670         }
 671
 672         bgp_path_info_mpath_count_set(dmed_best, 0);
 673         UNSET_FLAG(dmed_best->flags, BGP_PATH_MULTIPATH_CHG);
 674         assert(bgp_path_info_mpath_first(dmed_best) == 0);
 675 }
 676
 677 /*
 678  * bgp_path_info_mpath_aggregate_update
 679  *
 680  * Set the multipath aggregate attribute. We need to see if the
 681  * aggregate has changed and then set the ATTR_CHANGED flag on the
 682  * bestpath info so that a peer update will be generated. The
 683  * change is detected by generating the current attribute,
 684  * interning it, and then comparing the interned pointer with the
 685  * current value. We can skip this generate/compare step if there
 686  * is no change in multipath selection and no attribute change in
 687  * any multipath.
 688  */
 689 void bgp_path_info_mpath_aggregate_update(struct bgp_path_info *new_best,
 690                                           struct bgp_path_info *old_best)
 691 {
 692         struct bgp_path_info *mpinfo;
 693         struct aspath *aspath;
 694         struct aspath *asmerge;
 695         struct attr *new_attr, *old_attr;
 696         uint8_t origin;
 697         struct community *community, *commerge;
 698         struct ecommunity *ecomm, *ecommerge;
 699         struct lcommunity *lcomm, *lcommerge;
 700         struct attr attr = {0};
 701
 702         if (old_best && (old_best != new_best)
 703             && (old_attr = bgp_path_info_mpath_attr(old_best))) {
 704                 bgp_attr_unintern(&old_attr);
 705                 bgp_path_info_mpath_attr_set(old_best, NULL);
 706         }
 707
 708         if (!new_best)
 709                 return;
 710
 711         if (!bgp_path_info_mpath_count(new_best)) {
 712                 if ((new_attr = bgp_path_info_mpath_attr(new_best))) {
 713                         bgp_attr_unintern(&new_attr);
 714                         bgp_path_info_mpath_attr_set(new_best, NULL);
 715                         SET_FLAG(new_best->flags, BGP_PATH_ATTR_CHANGED);
 716                 }
 717                 return;
 718         }
 719
 720         bgp_attr_dup(&attr, new_best->attr);
 721
 722         if (new_best->peer && bgp_flag_check(new_best->peer->bgp,
 723                                              BGP_FLAG_MULTIPATH_RELAX_AS_SET)) {
 724
 725                 /* aggregate attribute from multipath constituents */
 726                 aspath = aspath_dup(attr.aspath);
 727                 origin = attr.origin;
 728                 community =
 729                         attr.community ? community_dup(attr.community) : NULL;
 730                 ecomm = (attr.ecommunity) ? ecommunity_dup(attr.ecommunity)
 731                                           : NULL;
 732                 lcomm = (attr.lcommunity) ? lcommunity_dup(attr.lcommunity)
 733                                           : NULL;
 734
 735                 for (mpinfo = bgp_path_info_mpath_first(new_best); mpinfo;
 736                      mpinfo = bgp_path_info_mpath_next(mpinfo)) {
 737                         asmerge =
 738                                 aspath_aggregate(aspath, mpinfo->attr->aspath);
 739                         aspath_free(aspath);
 740                         aspath = asmerge;
 741
 742                         if (origin < mpinfo->attr->origin)
 743                                 origin = mpinfo->attr->origin;
 744
 745                         if (mpinfo->attr->community) {
 746                                 if (community) {
 747                                         commerge = community_merge(
 748                                                 community,
 749                                                 mpinfo->attr->community);
 750                                         community =
 751                                                 community_uniq_sort(commerge);
 752                                         community_free(&commerge);
 753                                 } else
 754                                         community = community_dup(
 755                                                 mpinfo->attr->community);
 756                         }
 757
 758                         if (mpinfo->attr->ecommunity) {
 759                                 if (ecomm) {
 760                                         ecommerge = ecommunity_merge(
 761                                                 ecomm,
 762                                                 mpinfo->attr->ecommunity);
 763                                         ecomm = ecommunity_uniq_sort(ecommerge);
 764                                         ecommunity_free(&ecommerge);
 765                                 } else
 766                                         ecomm = ecommunity_dup(
 767                                                 mpinfo->attr->ecommunity);
 768                         }
 769                         if (mpinfo->attr->lcommunity) {
 770                                 if (lcomm) {
 771                                         lcommerge = lcommunity_merge(
 772                                                 lcomm,
 773                                                 mpinfo->attr->lcommunity);
 774                                         lcomm = lcommunity_uniq_sort(lcommerge);
 775                                         lcommunity_free(&lcommerge);
 776                                 } else
 777                                         lcomm = lcommunity_dup(
 778                                                 mpinfo->attr->lcommunity);
 779                         }
 780                 }
 781
 782                 attr.aspath = aspath;
 783                 attr.origin = origin;
 784                 if (community) {
 785                         attr.community = community;
 786                         attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_COMMUNITIES);
 787                 }
 788                 if (ecomm) {
 789                         attr.ecommunity = ecomm;
 790                         attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_EXT_COMMUNITIES);
 791                 }
 792                 if (lcomm) {
 793                         attr.lcommunity = lcomm;
 794                         attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_LARGE_COMMUNITIES);
 795                 }
 796
 797                 /* Zap multipath attr nexthop so we set nexthop to self */
 798                 attr.nexthop.s_addr = 0;
 799                 memset(&attr.mp_nexthop_global, 0, sizeof(struct in6_addr));
 800
 801                 /* TODO: should we set ATOMIC_AGGREGATE and AGGREGATOR? */
 802         }
 803
 804         new_attr = bgp_attr_intern(&attr);
 805
 806         if (new_attr != bgp_path_info_mpath_attr(new_best)) {
 807                 if ((old_attr = bgp_path_info_mpath_attr(new_best)))
 808                         bgp_attr_unintern(&old_attr);
 809                 bgp_path_info_mpath_attr_set(new_best, new_attr);
 810                 SET_FLAG(new_best->flags, BGP_PATH_ATTR_CHANGED);
 811         } else
 812                 bgp_attr_unintern(&new_attr);
 813 }